1.2 | Capabilities & Limitations: Where AI stands today

Here we take a close look at where AI tools can help in everyday work – as of July 2025.

Learning Objectives:

  • Understand which tools and AI agents are suitable for which tasks.
  • Recognize important criteria: context, logical reasoning, usability, and cost.
  • Make an informed decision rather than a purely marketing-driven one.
✓

Very Well Suited

★★★★★

AI tools and agents can now handle these activities very reliably:

  • Standardized communication & content creation (e-mails, blog drafts, social media posts).
  • High-quality image & video generation (Midjourney V7, Sora, Runway Gen-4).
  • Speech & data processing (real-time transcription, translation, classification, summaries).
  • Autonomous automation of routine tasks by AI agents (e.g., research and summarization, travel planning, data entry).
A final human review remains crucial for quality, fact-checking, and compliance with company policies.
i

Well Suited (with expert review)

★★★★

For these tasks, AI tools provide strong support but require expert review:

  • Complex analyses & reports (market analyses, process documentation, presentation creation).
  • Sophisticated technical translations & code generation.
  • Strategic creative work (ideation, design concepts, campaign drafts).
  • Data analysis & dashboards (especially with RAG for connecting to current company data).
Human expertise is essential for reviewing logic, strategic alignment, nuances, and facts.
!

Moderately to Challengingly Suited

★★★

These areas continue to require a high level of human competence and strategic judgment:

  • Deep causal understanding & problem-solving in entirely new situations (hallucinations remain a residual risk).
  • Emotional intelligence, empathy, and complex negotiation.
  • Legal & ethical responsibility (e.g., within the framework of the EU AI Act, whose core obligations have been in effect since August 2025).
  • Final strategic, disruptive, or business-critical decisions.
  • Interpretation of ambiguous, highly unstructured, or emotionally charged data.
AI results should be considered as well-founded suggestions, not infallible instructions.

Key AI Tools at a Glance (July 2025)

  • ChatGPT (OpenAI)
    For: All-around tasks, multimodal interaction, creative content, and autonomous agent tasks.
    Current model: GPT-5 is expected for summer 2025 and is intended to unify various specialized models (like GPT-4o for multimodality and o3 for logic).
    Strengths: Very intuitive, fast, and flexible; the new „Agent“ mode can independently perform multi-step tasks online.
    Weaknesses: Inaccuracies can occur in complex analyses; the quality is highly dependent on the underlying model.
  • Claude (Anthropic)
    For: Sophisticated text work, analysis of long documents, programming, reliable and secure applications.
    Current models: Claude 4 family (Opus 4 & Sonnet 4), released in May 2025.
    Context window: 200,000 tokens.
    Strengths: Excellent in logic, code generation, and creating nuanced, professional texts. Opus 4 is considered one of the leading models for complex programming tasks.
    Weaknesses: The context window is smaller than Gemini’s; sometimes a bit more conservative in creativity.
  • Gemini (Google)
    For: Analysis of extremely large datasets, multimodal tasks (text, code, image, video, audio), deep integration with Google Workspace & Cloud.
    Current model: Gemini 2.5 Pro / Flash-Lite (stable since July 2025).
    Context window: 1 million tokens by default, experimentally up to 2 million.
    Strengths: The huge context window allows for the analysis of entire codebases or extensive reports in one go. Strong integration into the Google ecosystem.
    Weaknesses: Processing very large contexts can increase response time and requires careful handling to ensure detailed accuracy.
  • Image & Video Generators (Midjourney, Sora, Runway)
    For: Creating high-quality images, marketing materials, storyboards, and short video clips.
    Current versions: Midjourney V7 (since Q1/Q2 2025), OpenAI Sora, Runway Gen-4.
    Strengths: Generate results that are partly photorealistic and artistically sophisticated with improved coherence and detail.
    Weaknesses: Requires practice in „prompting“; copyright and ethical issues are still under intense discussion.
  • GitHub Copilot / Amazon CodeWhisperer
    For: Autonomous programming support, code modernization, debugging, and test automation.
    Strengths: Development towards „AI Agents“ that can independently perform tasks like modernizing .NET applications. Deep IDE integration.
    Weaknesses: Generated code must be checked for security, efficiency, and hidden vulnerabilities. It does not replace understanding the codebase.
Note: The AI landscape is evolving rapidly. New tools and improved models are constantly entering the market. It is important to understand the fundamental capabilities and limitations.

We hope this module has given you a clear and up-to-date overview of the possibilities and limitations of AI in a work context. In the next module, we will discuss how to make optimal use of AI assistants through effective „prompting“.